The Atlantic publishes a searchable list of music used to train AI

In short: The Atlantic has made four large collections of music data searchable, so the public can see what songs may have been used to train AI music systems.

What happened

Atlantic reporter Alex Reisner found four datasets, meaning big collections of files or links used to train AI models (software that learns patterns from examples, like a student studying a huge stack of songs). Reisner then turned those datasets into a public, searchable database.

Two of the datasets are especially large, with about 12 million and 9 million tracks. The other two are smaller, but still contain more than 100,000 songs each.

Reisner says the datasets have been downloaded thousands of times. It is not possible to know exactly who has used them, but Google and Stability have confirmed in research papers that they used them.

The reporting also notes that three of the datasets are not shared as audio files. Instead, they are lists of links to songs on YouTube or Spotify. AI developers can use automated tools to download the audio in bulk. Reisner says some of these tools can bypass logins, ads, and other systems that help pay creators, and that this can violate the platforms’ rules.

The datasets include well known artists, including Lady Gaga, Radiohead, Aphex Twin, Wu-Tang Clan, and Bruce Springsteen.

Why it matters

For musicians and listeners, this database makes a usually hidden part of AI development easier to inspect. It can help artists check whether their work appears in common training sources, and it adds evidence to ongoing debates about permission, payment, and copyright when AI systems learn from existing music.

Source: The Verge AI